showstack

DMX refresh & what RDM costs

DMX refresh is not a setting, it is arithmetic. 250 kbit/s with eleven bits per slot is 44 µs a slot, and a full 513-slot frame plus break and mark comes to 22.7 ms — the familiar 44 Hz ceiling. The only lever is sending fewer slots. RDM shares the same pair by taking turns on it, so every transaction is time the transmitter is not sending levels: an RDM packet is 25 bytes before it carries anything, and a request plus a response plus two line turnarounds is a few milliseconds each. That is why heavy discovery or continuous sensor polling really does make a rig sluggish. How RDM answers back on a one-way wire →

Timecode

Drop-frame skips frame labels 00 and 01 at the start of every minute except each tenth minute — labels, not time. Entering a label that does not exist (say 00:01:00;00 in DF) is reported as such rather than silently rounded. See LTC and MTC.

MIDI hex decoder

One rule does all the framing: a status byte has its top bit set (80–FF), a data byte does not (00–7F). There is no header, no length field and no checksum, so a receiver joining mid-stream resynchronises on the next byte with the high bit set. For channel messages the high nibble is the command and the low nibble is the channel — zero-based on the wire, displayed one-based nearly everywhere, which is the off-by-one everybody meets once. Running status lets a repeated status byte be omitted, and it is why Note On at velocity 0 exists as a Note Off: a whole passage of notes shares one 9n byte. How MTC and MSC sit on top of this →

Relay logic matrix

Write each output as a boolean rule: & AND, | OR, ! NOT, parentheses group. Every input combination is evaluated into the matrix, which is how you sanity-check an interlock chain before wiring it. Up to 5 inputs and 6 rules; outputs cannot feed back, because latching and timing belong in the controller, not a truth table. This is a thinking tool: a real e-stop chain is hard-wired to the machinery standards, never through software.

Audio levels & impedance

dBu and dBV are both line-level voltage references, fixed 2.21 dB apart at any level — 0 dBu = 0.775 V RMS (the "unloaded" successor to 600 Ω-referenced dBm), 0 dBV = 1 V RMS. Pro gear nominally runs +4 dBu; consumer/semi-pro gear −10 dBV — an 11.8 dB gap, the reason a "line level" cable between the two clips or hisses until you pad or gain-stage it.

SPL and dBu/dBV are not the same kind of measurement and do not convert into each other: one is acoustic pressure in air, the other is electrical voltage in a cable. A mixer's output meter reading "0 dBu" says nothing about how loud the room is.

Ohms (Ω) also names two different things on this page. Resistance — the Ohm's law tool below, a lamp or heater element — opposes current the same way at any frequency, all of it dissipated as heat. Impedance — the Speaker load tool above — is resistance's AC generalisation, Z = R + jX: a reactance X from the driver's voice coil and crossover that shifts with frequency. A loudspeaker's "8 Ω" is a nominal average, not a fixed value — the real number can swing from under 5 Ω to well over 40 Ω near cone resonance. That is why the speaker load arithmetic above is exact for a stated nominal figure, while Ohm's law's resistive-only assumption is indicative, not exact, once it is pointed at a driver instead of a lamp.

Light and sound both obey the inverse square law because both radiate from a small source across an expanding sphere: double the distance and the energy spreads over 4× the area, so the level at any point is quartered. For light that is illuminance — lux = candela ÷ throw², see Beam & throw below — a fixture twice as far away lights its target at a quarter the lux, all else equal. For sound in a free field (no walls or ground reflection filling it back in) the same physics shows up as a level drop rather than a ratio: −6 dB every doubling of distance, +6 dB every halving. How to use it: to sanity-check a claimed SPL at FOH against a spec measured at 1 m, count doublings of distance and subtract 6 dB each — a source rated 100 dB at 1 m is roughly 88 dB by 4 m (two doublings) outdoors. Indoors, reflections refill part of that drop, so 6 dB/doubling is the conservative, worst-case figure for clearance and neighbour-noise planning, not what a meter will actually read in a live room.

dBu, dBV, volts and dBFS

Three things called dB that are not the same. dBu is referenced to 0.7746 V — the voltage that dissipates a milliwatt in 600 ohms, which is why the number is not round. dBV is referenced to 1 V, so the two differ by a constant 2.2 dB and convert exactly. dBFS converts to neither until somebody decides where the alignment level sits, and that decision is regional: Europe puts 0 dBu at −18 dBFS, the US puts +4 dBu at −20 dBFS. A feed arriving consistently 4 dB out between a European and an American facility is almost always this and not a fault.

Noise exposure dose

The rule selector picks the criterion level, the exchange rate and the criterion duration together. Exchange rate is the whole argument: 3 dB halves the permitted time for every 3 dB louder, 5 dB is far more permissive, and which one applies is a matter of jurisdiction, not physics. This is crew exposure under 2003/10/EC or OSHA 1910.95. Audience exposure is a separate question with its own document — in Germany, DIN 15905-5.

Room modes

Below the Schroeder frequency a room does not behave statistically — individual standing waves dominate and the response at a seat is set by the room’s dimensions rather than by the system. Absorption on the walls does not fix that; the wavelengths are metres long. Axial modes (two parallel surfaces) are the loud ones. Rooms whose dimensions are simple multiples stack their modes instead of spreading them, which is why a cube sounds bad. Measuring and aligning sound has the RT60 side.

Line array coverage

A point source loses 6 dB per doubling of distance because the energy spreads over a sphere. A line source long compared with the wavelength spreads cylindrically and loses 3 dB — which is the whole reason an array reaches the back without removing the front row’s hearing. It does not last: beyond a transition set by array length squared and frequency, the wavefront becomes spherical again. Designing as though the 3 dB region reached the back wall is the classic mistake.

What a waveform is made of

Every periodic wave is a sum of sines at multiples of its fundamental, and the four classic shapes have tidy recipes. A sawtooth has every harmonic at 1/n, which is why it is bright. A square has odd harmonics only, so it is missing all the octaves and sounds hollow. A triangle is odd harmonics at 1/n², falling so fast that the third is already down at a ninth, which is why it is nearly a sine. The odd-only shapes put a lot on the third harmonic — which is a triplen, the same one that fills a three-phase neutral. An oscillator making a square wave and a switch-mode supply drawing a spiky current are the same fact pointed at different jobs.

Amplitude panning: VBAP and DBAP

VBAP in its two-speaker form is the tangent law, which is what a pan pot has approximated since before anybody wrote it down. Gains are normalised for constant power, not voltage, which is why dead centre is −3 dB in each speaker and not −6. DBAP asks a different question: it does not assume where the listener is, only how far each speaker is from the position the source should occupy. The rolloff exponent is a design choice rather than physics — high localises hard and jumps audibly between speakers, low smears the image across the rig and survives an audience that is not in one seat. That trade is the whole reason to choose one over the other.

LED wall pitch and viewing distance

The retina distance is where one pixel subtends one arcminute, which is the same acuity figure behind legible text on the senses page. It is a threshold rather than a rule: the front row is routinely inside it, and the content decides whether that matters. Hard-edged graphics and small text show the grid a long way before a soft image does. The data rate is raw pixels at 24-bit colour and 60 fps before any transport overhead, which is the number that decides how many processor outputs the wall needs.

Aspect fit

Fit letterboxes and wastes surface; fill crops and loses content. The bar and crop figures are the numbers a designer needs, because that is the dead area to mask or design around. Scaling past 1:1 is flagged separately — that is the point where an LED wall starts to look soft, and no amount of processing puts the pixels back.

Colour mixing & shadows

Two operations share the word mixing and they are opposites. Additive is emitters: each source contributes its own spectrum and they sum, so you start at black and add, and red plus green is yellow. Subtractive is gel, CMY flags and costume: each layer removes part of the spectrum, which is multiplication, so you start at white and take away, and a deep colour is always a dim one. All of it is computed in linear light, not in code values — two sources at 50% make 176, not 255. The shadow swatches are the answer to why shadows go coloured: a shadow is the light that still arrives, so blocking one of two sources leaves the other one’s colour. The whole mechanism →

Colour temperature correction

Kelvin is the wrong scale for this arithmetic: the step from 3200 K to 3400 K looks far bigger than the step from 9000 K to 9200 K, so no gel can have a fixed effect stated in kelvin. The mired — micro reciprocal degree, 10⁶/K — is the scale on which correction is fixed, which is why every swatch book prints a mired shift. Gel values are the published Lee shifts; Rosco equivalents differ slightly. How a colour becomes a number covers why.

Cable derating & gauge

A cable’s rating is measured in still 30 °C air with three current-carrying conductors, and a touring rig gives it none of those things. Two published factors correct for it: bundling from NEC Table 310.15(C)(1), and ambient temperature from Table 310.15(B)(1) — computed here from the formula that table was generated from, √((Tc−Ta)/(Tc−30)), so it lands on the published values and keeps working between the rows. The base rating has to come off your cable’s own datasheet; a tool that guessed would be worse than no tool. Then check the run with voltage drop — on a long run that, not heat, is usually what sizes the cable.

Harmonics & the neutral

A perfect load draws a sine. Nothing on a rig does — switch-mode supplies draw in spikes near the voltage peak, and a spiky current is a sine plus harmonics at multiples of the mains frequency. On three phases the fundamentals cancel in the neutral, but the triplens (3rd, 9th, 15th) arrive in phase on all three legs and add there instead. So a perfectly balanced rig can put more current down the neutral than any phase is carrying, and the neutral is the one conductor with no breaker in it. THD-F and THD-R are both in use and are different numbers, which is why a meter and a datasheet can disagree while both are right. Why three phases cancel, and what stops them →

Bridle angle — why it is never half each

This is a geometry explainer, not a design tool. It shows one symmetric two-leg bridle with the load hanging at the apex, and nothing else: no sling angle derating, no shock load, no self-weight, no assessment of whether the structure can take the sideways pull it shows you. Real bridles are designed by a qualified rigger against DIN 56950-1, EN 17206 or the local equivalent. What it is good for is the thing people get wrong from memory: tension per leg is W / (2 cos θ), so at 60° from vertical each leg is carrying the whole load, not half of it.

Wind load on a surface

A screening number, not a design. Force goes with the square of wind speed, so doubling the wind quadruples the load — that is the whole reason a wind action plan carries speeds rather than opinions. What this does not model: terrain and height factors, the difference between a 10-minute mean and a 3-second gust at your actual site, dynamic response, or anything about how the structure is built. Temporary demountable structures are designed to EN 13782 or ANSI E1.21 by somebody competent to do it. Use this to decide whether to have the conversation.

Pepper’s ghost contrast

Everybody gets the geometry right and the contrast wrong. The image sits as far behind the pane as the object is in front — that part is easy. What decides whether it reads as a solid figure or a smear on a window is two luminances arriving at the same retina: the object times the pane’s reflectance, against the set behind it times the pane’s transmittance. There are only two moves, and brightening the object is the expensive one. Darkening what sits behind the ghost is almost always cheaper, and it is the one people reach for last. The rest of the craft →

Forced perspective

Two things look the same size when they subtend the same angle, and angle is size over distance — so an object twice as far away has to be twice as big. That is the whole trick, and the arithmetic is the easy half. The useful half is where it stops: angular size is one depth cue among several, and inside about ten metres binocular disparity simply overrules it. A forced-perspective set that is perfect in a photograph collapses for the front row, because a camera has one eye and an audience has two. Motion parallax does the same job for anybody who moves their head. The technique is really a statement about who is allowed to look, and from where.

Flash rate & photosensitivity

Three flashes in any one second is the line, and it is the same number in WCAG 2.3.1, in ITU-R BT.1702 and in the Ofcom guidance. Sensitivity peaks between 15 and 20 Hz, which is exactly where a strobe lands when somebody sets it by ear. Saturated deep red is judged more strictly than any other colour, and static patterns count too: more than five clearly discernible light-dark stripe pairs is its own hazard. This is a screening check — broadcast material is assessed with a Harding test, and a venue still needs signage at the door.

Subnet calculator

The mask says how many of the 32 bits are the network; everything else follows from that. A /31 is a point-to-point link with both addresses usable (RFC 3021) and a /32 is a single host, which is why neither reserves a broadcast address. How to calculate it by hand →

What kind of address is this?

Every reserved range carries a diagnosis. 169.254.x.x means DHCP failed and the device named itself — it can reach others that did the same, on the same wire, and nothing else. 127.x never leaves the machine, so a service bound only there is unreachable however good the cable is. 239.255.x.x is an sACN group rather than a host, and needs IGMP snooping to behave. And the private ranges are private because routers on the public internet drop them — which is why a show network lives in one, and why that is a security property you get for free by choosing the right numbers.

Terminal commands for a show network

Nearly every network question on a show has a one-line answer from a terminal, and the reason people do not use them is that the command has a different name on every platform and the output is unlabelled. The third column is the part a cheat sheet usually leaves out: what in the output actually answers the question. Two of these are worth learning before the others — the route table, because a laptop with both wifi and a show network has two default routes and traffic vanishes into the wrong one; and the multicast joins, because a receiver that has not joined the group will never see it however correctly you send.

Power over Ethernet budget

Two numbers, and confusing them is how a rig of cameras browns out on day two. The PSE figure is what the switch port may put on the cable; the PD figure is what the standard guarantees at the far end. The gap between them is the cable, sized by IEEE for 100 m of the worst compliant cable — so the guarantee is a floor, not a prediction, and real cable does much better. CCA is not a cost saving: about 60% more resistance than copper, which turns a working bench test into a brownout on a long run. How power gets to the thing →

PoE switch budget

The per-port maximum and the whole-switch budget are separate limits, and passing one says nothing about the other. A 24-port switch advertising PoE+ on every port almost never has a supply that can do 24 × 30 W. When it runs out it sheds ports by priority rather than derating them all — so the question is not whether it copes but which ports you have decided may fail. Set that deliberately, or the switch decides for you on the show day.

Fibre loss budget

Attenuation figures are typical values per TIA-568 and FOA guidance: OM3/OM4 about 3.0 dB/km at 850 nm and 1.0 at 1300; OS2 about 0.4 at 1310 and 0.3 at 1550. Connector pairs are counted at 0.3 dB (TIA allows up to 0.75) and fusion splices at 0.1. The link budget belongs to the optics, not the glass — take it from the transceiver datasheet rather than this default. A run that passes with under 3 dB spare works on the day and fails after one re-terminated connector.

SDI reach on coax

SDI does not degrade — it works perfectly and then stops, which is why a run that was fine in the shop fails in the venue ten metres longer. Two facts set the cliff: coax loss rises with the square root of frequency, and the frequency that matters is half the bit rate. Take both cable numbers off the manufacturer’s datasheet, at whatever frequency they quoted; every coax maker publishes them. The 20 dB equalisation figure is what SMPTE writes down, and real receivers often do better, which is why the same cable gets quoted at different lengths by different people. A run inside 3 dB of the budget is flagged: it works today and fails after somebody swaps a barrel in.

Why a Blu-ray holds more than a CD

Same 120 mm disc, same spin, same idea — a spiral of pits read by a laser. What changed is how small a spot the optics can focus, because that sets how small a pit can be and how close two tracks can run. A focused spot cannot beat diffraction: 1.22 λ / NA. Infrared to red to blue-violet, and 0.45 to 0.85, takes the spot from about 2.1 µm to 0.58 µm — and density goes with the square of that, because a disc is a surface. Capacity then outruns the area ratio, and that gap is the coding: better modulation and better error correction, on top of what the physics gave.

RC time constant & filter corner

One resistor and one capacitor is the most useful circuit in the trade. It is a filter, a delay, a de-bounce, and the tone control on every analogue console ever built — and it is two equations that are the same fact: τ = RC and f = 1 / 2πRC. The corner is where the capacitor’s reactance equals the resistance. One pole is 3 dB down there and 6 dB per octave after, and every extra stage adds another 6. In the time domain the same numbers say 63% of the way in one time constant, 95% in three, 99% in five — which is how long a mute ramp or a de-bounce actually takes.

OSC message

Four rules and everything follows. Strings are null-terminated then padded with more nulls to a multiple of four. Numbers are big-endian, four bytes. The type tag string starts with a comma and names each argument in order. And the whole message is always a multiple of four bytes — which is why an OSC dump is full of trailing zeros that mean nothing. Integers are sent as i, anything with a decimal point as f, everything else as s. Note that T, F, N and I carry a tag and no bytes at all.

PJLink projector command

TCP port 4352, one line per command: per cent, class, four upper-case letters, space, parameter, carriage return. On connect the projector greets you with PJLINK 0 (no security) or PJLINK 1 <8 hex digits> — paste those digits and your password above and the MD5 digest is prepended to the first command only, with no separator. Nothing typed here leaves your device; the password is used to compute a hash locally and is never sent anywhere or stored.

Art-Net packet

Eighteen header bytes and then the slots. Two things trip people: the opcode is sent low byte first while the data length in the same header is sent high byte first, which is a real inconsistency in the protocol rather than a mistake in your reading. And the sequence number is not a counter of slots — it is 1 to 255 with 0 meaning sequencing is off, used by a receiver to drop packets that overtook each other. ArtPoll is the broadcast that makes every node reply with ArtPollReply, which is how a controller builds its node list.

sACN packet

Three nested PDUs — root, framing, DMP — each opening with a combined flags-and-length field where the top nibble is 0x7 and the remaining twelve bits are that PDU’s own length. Getting one of the three wrong produces a packet some receivers accept and others silently drop, which is a miserable fault to chase. A full universe is 638 bytes: root 622, framing 600, DMP 523. The CID identifies the source, not the universe, and must not change between packets — a device generating a fresh one each time looks like unlimited new sources and breaks priority arbitration.

RDM packet

Twenty-four header bytes, then parameter data, then a two-byte additive checksum — a plain sum of every preceding byte, which catches exactly the single-bit errors a marginal RS-485 line produces. The message length field counts everything up to but not including the checksum, which is the off-by-two everybody hits once. Start code 0xCC is what tells a fixture this is management traffic rather than levels. How RDM answers back on a one-way wire →

MIDI Show Control & Machine Control

The one thing on this page that can actually leave the machine. Browsers implement Web MIDI, so with a MIDI interface attached these bytes can be transmitted for real — your browser will ask permission first, including for system exclusive, which both of these are. Everything stays local: no network, nothing recorded. MSC cue data is ASCII digits, so cue 12 is 31 32 and not 0C, which is why a cue number can contain a decimal point. A receiver acts only if both its device ID and its command format match, and a format mismatch is the commonest reason MSC appears to do nothing at all.

Wake-on-LAN

Six FF bytes then the target MAC repeated sixteen times. That is the entire format — there is no checksum, no reply and no acknowledgement. Send it to the broadcast address of the target subnet rather than to the machine, because a machine that is off has no IP address to be addressed by. It will not cross a router unless directed broadcast is enabled, which it usually is not, so send it from the same VLAN. Port 9 is conventional and 7 is just as common; the network card watches for the byte pattern and does not care which port carried it.

VISCA over IP camera command

An eight-byte header in front of an ordinary VISCA message: payload type, payload length, and a sequence number the camera echoes back. The VISCA message itself is unchanged from the serial version, which is why a serial-to-IP converter is only wrapping and unwrapping this header. The address byte is 0x80 plus the camera number, which is why every VISCA example starts with 81 and never explains it. Sequence numbers matter: some cameras ignore a command whose number has gone backwards, which is why two controllers on one camera behave oddly rather than simply clashing.

DDP pixel packet

A ten-byte header then pixel data, three bytes per pixel. The push bit is the one that catches people: with it clear the controller stores the data and displays nothing, so a first DDP test appears to do absolutely nothing while being entirely correct. Offset is in bytes, not pixels, so the second pixel starts at offset 3. One packet carries up to 1440 bytes of data, which is 480 pixels; a longer strip is several packets at increasing offsets with push set only on the last.

Third-order intermod check

Two transmitters make products at 2a−b and 2b−a; three make a+b−c. Third order is the set that matters because the products land near the originals and are strong enough to open a receiver. A product in empty spectrum is harmless — the ones flagged in red are landing on a channel you are actually using. This is a check, not a coordination tool: it ignores transmitter power, antenna placement, receiver selectivity, fifth-order products and every broadcaster already on air, which is what the frequency map and a real coordination pass are for.